TREATMENT HETEROGENEITY AND POTENTIAL OUTCOMES IN LINEAR MIXED EFFECTS MODELS

摘要: Studies commonly focus on estimating a mean treatment effect in population. However, some applications the variability of effects across individual units may help to characterize overall Consider set treatments, {T,C}, where T denotes that might be applied an experimental unit and C control. For each units, duplet { , }, represents potential response th if were control applied, respectively. The causal compared is difference between two responses, . Much work has been done elucidate statistical properties effect, given particular assumptions. Gadbury others have reported this for simple designs primarily focused finite population randomization based inference. When become more complicated, approach becomes increasingly difficult. Since linear mixed models are particularly useful modeling data from complex designs, their role heterogeneity investigated. It shown can conceptualized as combination fixed random effects. assumed variance components specified ―potential outcomes‖ model when both outcomes, variables model. used quantify heterogeneity. Post assignment, however, only one outcomes observable unit. then component non-estimable analysis observed data. Furthermore, estimable demonstrated arise combinations Mixed considered context design effort illuminate loss information incurred moving framework analysis. TREATMENT HETEROGENEITY AND POTENTIAL OUTCOMES IN LINEAR MIXED EFFECTS MODELS